diff --git a/CHANGELOG.md b/CHANGELOG.md
index da2d5e260..cd505abba 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -80,6 +80,15 @@ Open cycle — bug-fix / housekeeping. Entries land here as they merge.
   large table this removes the dominant per-cell layout allocation. No public API
   or behaviour change.
 
+- **Process-wide line-metrics cache stops inserting instead of flushing when full.**
+  The static line-metrics cache `clear()`-ed every entry once it passed 50,000
+  distinct styles — a full flush whose non-atomic check-then-clear is a
+  thundering-herd recompute under concurrent rendering. It now stops inserting at
+  the cap and keeps the existing entries (distinct styles are few in real use, so
+  this is only a pathological-explosion guard; it runs on a cache miss, never on
+  the per-measurement path). **Measured line metrics are unchanged.** No public API
+  or behaviour change.
+
 ### Tests / tooling
 
 - **Benchmark regression gate and measurement probe (benchmarks module, not part
diff --git a/src/main/java/com/demcha/compose/engine/measurement/FontLibraryTextMeasurementSystem.java b/src/main/java/com/demcha/compose/engine/measurement/FontLibraryTextMeasurementSystem.java
index b94e3f29b..500e219d5 100644
--- a/src/main/java/com/demcha/compose/engine/measurement/FontLibraryTextMeasurementSystem.java
+++ b/src/main/java/com/demcha/compose/engine/measurement/FontLibraryTextMeasurementSystem.java
@@ -100,10 +100,16 @@ private GlobalPdfStyleKey globalPdfStyleKey(PdfFont font, TextStyle style) {
     }
 
     private static void cacheGlobalLineMetrics(GlobalPdfStyleKey key, LineMetrics metrics) {
-        if (GLOBAL_PDF_LINE_METRICS_CACHE.size() > GLOBAL_LINE_METRICS_CACHE_LIMIT) {
-            GLOBAL_PDF_LINE_METRICS_CACHE.clear();
+        // Safety cap on the process-wide line-metrics cache. Distinct styles are
+        // few in real use (a handful of font/size/decoration combos); this only
+        // guards a pathological style explosion. Stop inserting once full instead
+        // of clear()-ing: the old full flush wiped every hot entry under
+        // concurrent rendering (a thundering-herd recompute), so keeping the
+        // existing entries is strictly better. This runs on a cache miss only,
+        // never on the per-measurement get() path.
+        if (GLOBAL_PDF_LINE_METRICS_CACHE.size() < GLOBAL_LINE_METRICS_CACHE_LIMIT) {
+            GLOBAL_PDF_LINE_METRICS_CACHE.putIfAbsent(key, metrics);
         }
-        GLOBAL_PDF_LINE_METRICS_CACHE.putIfAbsent(key, metrics);
     }
 
     @Override